Tagging Continuous Labels for EEG-based Emotion Classification.

Area of Science:

• Neuroscience
• Affective Computing
• Machine Learning

Background:

• Electroencephalography (EEG)-based emotion classification is crucial for affective brain-computer interfaces (aBCIs).
• Current methods predominantly use discrete labels, which oversimplify continuous emotional states and may introduce labeling errors.
• This imprecision in discrete labeling can impede advancements in emotion recognition.

Purpose of the Study:

• To develop an efficient system for continuous labeling of EEG data.
• To construct a novel EEG emotion dataset with continuous labels.
• To demonstrate the superiority of continuous labeling over discrete methods for emotion classification.

Main Methods:

• Developed a system to assign a unique, continuous label to each EEG sample.
• Created a continuously labeled EEG emotion dataset.
• Experimented with various classification models using both continuous and discrete labels.

Main Results:

• Continuous labeling significantly improved emotion classification performance compared to discrete labeling.
• Identified EEG features related to induced and non-induced emotions using continuous labels.
• Demonstrated the learnability and generalizability of EEG features with continuous labels across datasets.

Conclusions:

• Continuous labeling offers a more accurate approach to EEG-based emotion classification.
• The developed dataset and methods provide a valuable resource for aBCI research.
• Continuous labels enhance the understanding of EEG dynamics in emotional states.